ONC HTI-2: Unpacking AI Transparency for Cardiac Health Investors

The regulatory currents governing artificial intelligence in healthcare are shifting, and the ONC’s earlier HTI-2 proposals for algorithm transparency represented a significant inflection point, particularly for developers of predictive decision support algorithms. These proposals aimed to mandate a deeper, more granular level of transparency, fundamentally altering the compliance field. For federal policy analysts and health system legal counsel, understanding these expanded obligations as originally conceived is paramount to effectively monitor compliance and navigate the evolving legal terrain.

From HTI-1 to HTI-2: A Sea change in Transparency

The initial Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Sharing (HTI-1) rule laid foundational groundwork for transparency in certified health IT. Its primary focus was on ensuring that health IT developers provided basic information about their algorithms. However, the ONC’s initial HTI-2 proposals elevated these expectations considerably, moving beyond general disclosures to demand detailed insights into the inner workings and potential impacts of predictive algorithms. This envisioned shift was not merely an incremental update. It signaled a regulatory intent to scrutinize the efficacy, fairness, and safety of AI-driven tools embedded within clinical workflows. The Department of Health and Human Services (HHS) and the Office of the National Coordinator for Health Information Technology (ONC) are responding to growing concerns about algorithmic bias, explainability, and the potential for these tools to exacerbate health disparities. The initial HTI-2 framework was conceived as a direct consequence of this evolving understanding, aiming to equip healthcare providers and, by extension, patients, with the necessary context to critically evaluate and appropriately use AI-powered decision support.

Expanded Disclosure Requirements for Predictive Algorithms

Under the ONC’s initial HTI-2 proposals, the data elements required for algorithm transparency were envisioned as significantly more extensive than those under HTI-1. Developers of predictive decision support tools would have faced obligations to disclose a complete array of information. This would have included, but would not have been limited to:

• Algorithm Design and Purpose: Detailed descriptions of the algorithm’s intended use, its underlying methodology, and the specific clinical problem it aims to address.
• Training Data Characteristics: Information about the datasets used to train and validate the algorithm, including demographic representation, data sources, and any known biases within these datasets. This is important for understanding potential algorithmic drift over time FDA guidance on AI/ML medical device change control.
• Performance Metrics: Clear and interpretable metrics of the algorithm’s performance, such as accuracy, precision, recall, and F1-score, broken down by relevant demographic subgroups to assess fairness and equity.
• Limitations and Risks: Explicit disclosures of the algorithm’s known limitations, potential failure modes, and any identified risks to patient safety or health equity.
• Update and Maintenance Procedures: A description of the processes for monitoring, updating, and maintaining the algorithm, including how algorithmic drift will be managed. This speaks directly to the need for a strong QMS / ISO 13485 for ongoing oversight.
• Human Oversight and Intervention: Guidance on the level of human oversight required, including when and how clinicians should override or interpret the algorithm’s recommendations.

These detailed requirements were intended to move beyond black-box disclosures, providing a more transparent view into how predictive models function and their potential implications. For instance, a major EHR vendor like Oracle Health, which implements numerous EHR-based predictive models, would have needed to ensure that their certified health IT products can capture and present this granular information effectively. The burden of compliance was envisioned to shift from a superficial “what does it do?” to a complete “how does it work, for whom, and under what conditions?”

Compliance Deadlines and the Urgency of Adaptation

While the ONC HTI-2 has been partially finalized, the complete algorithm transparency requirements discussed here were part of earlier proposals and not enacted in their entirety. The original proposed timeline suggested a phased implementation, but the core requirements for enhanced transparency as initially envisioned were expected to take effect swiftly. This means federal policy analysts were preparing to evaluate compliance against these new, more stringent criteria, and health system legal counsel were advising their organizations on the necessary adjustments to procurement, implementation, and oversight of predictive algorithms. The urgency is amplified by the broader regulatory environment. The ECRI AI healthcare hazard rankings, for example, consistently highlight the risks associated with opaque or poorly understood AI in clinical settings. Similarly, the AMA’s legislative activity regarding AI healthcare oversight in 2026 shows a growing demand for accountability and ethical deployment. The initial HTI-2 proposals were a direct response to these concerns, aiming to provide a standardized framework for addressing them.

Implications for Developers and Health Systems

For health IT developers, the initial HTI-2 proposals meant a significant investment in internal processes, documentation, and potentially, redesign of their AI models to ensure transparency by design. Companies that have historically operated with less stringent disclosure practices would have needed to overhaul their approach. Those with a strong GMLP (Good Machine Learning Practice) foundation and a well-defined QMS will be better positioned to meet these new demands. For health systems, the implications of these proposals were equally deep. Legal counsel would have needed to review existing contracts with health IT vendors to ensure they align with the new transparency requirements. Plus, health systems would have gained access to more strong information about the predictive algorithms they deploy, enabling more informed decision-making regarding their use, potential biases, and impact on patient care. This enhanced transparency would also have aided in mitigating risks associated with algorithmic bias and improving patient safety, aligning with the principles of Real-World Evidence (RWE) in evaluating AI performance.

“The shift from HTI-1 to the initial HTI-2 proposals is not merely an update. It’s a recalibration of regulatory expectations for predictive decision support. This demands a proactive stance from both developers and implementers of healthcare AI.”, Healthcare AI Compliance Watch Editorial Board

Methodology and Source Note

This analysis is grounded in a comparative evaluation of the ONC HTI-1 framework and the ONC’s initial HTI-2 proposals. Our approach involved a detailed statutory and regulatory text comparison, drawing insights directly from the publicly available ONC HTI-2 proposed rule text and relevant Department of Health and Human Services (HHS) press releases. The information presented herein reflects our editorial board’s interpretation of these regulatory documents, focusing on their implications for healthcare AI regulatory compliance, particularly concerning the ECRI AI healthcare hazard and AMA AI healthcare oversight discussions anticipated for 2026. ONC HTI-2 proposed rule text HHS press releases on health IT initiatives The ONC’s initial HTI-2 proposals heralded a new era of transparency for predictive algorithms in healthcare. This heightened scrutiny, while challenging, was seen as essential for fostering trust, ensuring equity, and in the end, realizing the full potential of AI to improve patient outcomes responsibly. For federal policy analysts and health system legal counsel, staying abreast of these changes was not just a matter of compliance, but a critical component of strategic planning in the rapidly evolving field of healthcare AI.